Brain Cortical Surface Can Predict Fiber Trajectories

2179 

Abstract Submission 

Bao Ge¹

¹Shaanxi Normal University, Xi'an, China

Bao Ge  
Shaanxi Normal University
Xi'an, China

The brain primarily consists of gray matter and white matter, which interact with each other to form the complex functions of the brain. Research concerning the interplay between gray and white matter structures have offered some evidences support a link between the formation of cortical convolutions and the connectivity of underlying white matter [1]. Moreover, studies of brain diseases have found white matter network reconfiguration is a response to deep gray matter pathology in clinically isolated syndrome and early-stage relapsing-remitting multiple sclerosis patients [2]. These studies have revealed a clear correlation between gray matter and white matter.
This work tries to explore their close relationships from a novel perspective, which directly utilizes brain cortical surface information to predict the fiber trajectories. Previous research [3], focused on predicting the presence of fiber connections between two points on the brain surface. In contrast, this work advances this understanding by predicting the coordinate information of fiber trajectory between two points on the brain surface.

The overall framework of this work is depicted in Fig.1. The T1 MRI images and diffusion MRI (dMRI) images from 20 healthy subjects in the publicly available HCP S900 dataset [4] were used, the probabilistic fiber tracking were performed by MRtrix3, thus 10^6 fiber streamlines were obtained for each subject and the inner cortical surfaces were reconstructed by FreeSurfer.

Then we need to prepare sample data from actual data to train model parameters. Firstly, we selected those fibers with both endpoints located within the gray matter cortex. Then, we sampled fiber streamlines to standardize their length to 100 points for the purpose of simplifying the training process and facilitating evaluation. We selected the first and last points of each fiber streamline as part1 of training samples, which are also two cortical surface points. The remaining 98 points on the fiber which are to be predicted is part3 of training samples. The uppermost, lowermost, leftmost, rightmost, frontmost, rearmost points on the brain surface, as well as the geometric center point of the entire brain surface is part2 of training samples. We used part1 and part2 to predict part3.

We improved the BiLSTM model [5] for the regression task of predicting fiber coordinates. To gauge the proximity of the predicted fiber coordinates to the true fiber coordinates, we utilized evaluation metrics including Mean Squared Distance (MSD), Cosine Similarity, and Hausdorff distance. Among them, the MSD loss and cosine similarity loss function were also used in the training process.

We performed a K (K=5) folds cross-validation on a set of 20 individuals. We select the true representative fibers and their prediction results to show in Figure2. The true representative fibers are selected from the TractSeg fiber tract atlas [6] for each subject because we used the same subjects as the TractSeg dataset. It can be seen that the predicted fiber streamlines is very close to the true fiber streamlines. Additionally, we conducted quantitative evaluation experiments using data from different individuals, and computed the above three metrics. The average MSD, cosine similarity and Hausdorff distance are 21.5518, 0.9982 and 9.8085, respectively. Both qualitative analysis and extensive quantitative experiments confirm that information derived from the brain's cortical surface can be employed to predict fiber trajectories.

This study tried to use the cortical information to predict the trajectories of white matter fiber, which provides a new perspective for further study of the close relationship between gray matter cortex and nerve fibers.

Normal Brain Development: Fetus to Adolescence

Diffusion MRI Modeling and Analysis

Normal Development

White Matter Anatomy, Fiber Pathways and Connectivity ¹

Diffusion MRI ²

STRUCTURAL MRI

Sub-Cortical

White Matter

WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC